Reborn in 2010, the AI ​​in my mind is too powerful.

After returning home, Chen Yiran did not start working immediately. He needed AMD to send over the instruction set manual and the memory controller white paper first.

However, not everyone on AMD's side supports its actions.

The next day, in a small conference room at AMD, the four managers were having a small meeting.

Thomas Seifert, who sits at the head of the table, is the acting CEO and CFO. In fact, he made it clear from the first day he took office that he did not want to stay in the position for long, so he basically does not offer any opinions now and just wants to smoothly transition to the new CEO.

Next up was a portly man slumped in a chair. This was Rick Bergman, Senior Vice President and General Manager of the Product Group, responsible for all CPU and GPU product lines. Although not from a technical background, he was now the de facto number one figure in AMD's technical sphere.

Next to him was a tall, thin man with glasses. He was Eric Demers, a veteran from the ATI era, and currently the Chief Technology Officer of the graphics card division. He was the most influential person in the company when it came to graphics card technology.

Finally, there's Senior Vice President Harry, who has been communicating with Chen Yiran.

The four people remained silent, each engrossed in their phones, creating an extremely awkward atmosphere.

This is indeed the current situation at AMD: the acting CEO is not doing any work, and Harry and Bergman have some ideas, so they brought Chen Yiran into the company.

However, this drew strong dissatisfaction from Demers, and the entire graphics card division became very restless.

Just then, there was a knock on the door, and Marvin pushed it open and walked in, holding four printed PPT files in his hand.

As the leader of the technical team who communicates directly with Chen Yiran, he is going to give a report today.

After distributing the documents, Marvin began his tea ceremony performance.

"Gentlemen, let me give you a brief report on my initial discussion with Chen Yiran today."

Even if Seifert didn't want to get involved, he had to speak up now, because the other three were just staring at him like they were mute.

"Okay, tell us about the results of your conversation with Chen Yiran. Do you have any direction in mind?"

Marvin nodded and began to explain: "He has some different views on the architecture of our OpenCL compiler team."

He believes that the existing AMD IL middleware layer is too bloated, and that packing excessively long instructions at the end of the pipeline leads to the loss of a lot of scheduling information.

He proposed a redesign approach starting from the middle, with the overall idea being to move instruction dependency analysis to the generation stage.

Eric scoffed. "He dares to dream of it. Does he think it's something simple?"

Marvin said with a helpless expression, "Mr. Chen's work style is very confident. He reacts quickly to technical choices and the existing working methods of the entire team."

Well, how should I put it? There are some understandable reservations. Our current compiler kernel is a continuation of the ATI era, and Mr. Chen directly pointed out many legacy issues during the exchange.

Ignoring Eric, who was puffing out his beard and glaring at him, Bergman glanced at the document in his hand and said, "Your writing isn't very detailed either. You tell me his opinion."

"He believes that only half of the more than four hundred opcodes are actually used, while the other half consumes too much compilation time and maintenance costs. He thinks these things will hinder subsequent changes. Therefore, he doesn't want to optimize or rectify them; he wants to rewrite them completely."

The four people below visibly changed their expressions. Although they knew their skills weren't great, being completely dismissed was still quite upsetting.

Seeing the expressions on the faces of the others, Marvin continued to complain: "There's another situation. As we all know, Mr. Chen's first project was in the field of compilers, and he does have some insights in this area."

However, as everyone knows, GPU backend compilation is completely different, and he might not be very familiar with the hardware manual. My team and I suggested that if we went down the rewrite route, the hardware adaptation part might take a longer period of time.

We need to give Mr. Chen more support.

The four people below are not fools; Marvin, the old fox, is implying that Chen Yiran can't handle it.

Harry finished his self-planning first, then waved his hand: "Okay. You can go now."

After the door closed, Eric immediately spoke up: "I said from the beginning that this was a bad idea. I admit that what we inherited from ATI was not perfect, and the four architectural changes in the middle were even more chaotic."

But this is, after all, the culmination of decades of hard work; at least it's usable. You think just anyone under twenty can optimize it? And you even have the audacity to say you'll scrap it and rewrite it!

Harry shook his head and said, "If you're still questioning Chen Yiran's technical skills, then I think you might as well throw your brain away. If Chen Yiran says it needs to be rewritten, then I think it's you who needs to rewrite it."

Eric was still unconvinced: "I know what you mean. A $200 million compiler, DeepFlow, the acclaim from various companies, the so-called father of AI, and Jobs' $30 billion endorsement."

But have you ever considered this question: Was his previous product success due to superior technology, or the superior skills of the people in his company?

Harry laughed: "First of all, the compiler and Deepflow were both written by Chen Yiran as an independent creator, and I've told you before why I wanted to bring Chen Yiran into the company."

Because I have people working at Nvidia, although I dare not disclose specific technical data, I know a lot of administrative information. Including their private evaluations of Chen Yiran, which are very valuable pieces of information.

At Nvidia, many technical staff already treat Chen Yiran like a god. In their words, their communication with FlowEDA presents two very different experiences. One is the normal, collaborative discussion to resolve issues.

One type is clearly instructional: when Chen Yiran is at work, he can solve all their problems in five minutes.

Bergman nodded and said, "You all saw Nvidia's reaction after we officially announced our partnership with Chen Yiran last week. That day, Nvidia bought out all the McDonald's restaurants around here."

Three departments were working overtime in meetings until 3 a.m. If Chen Yiran is a fraud, what are they afraid of? I agree to continue supporting Chen Yiran. To put it bluntly, we have no market right now. Trying is better than doing nothing.

Eric didn't answer immediately. He knew in his heart that Chen Yiran was unlikely to be a fraud, but he was still unwilling to accept that his years of hard work had been overturned by a young man who was not even twenty years old.

They don't give us any face at all. If they had just come over and had a proper chat, that would have been great.

Seeing that Eric remained silent, Bergman became even more blunt: "Eric, I'm not arguing with you about specific technical details. I just want results. Can you handle our current OpenCL compiler? How competitive will it be? Can you give me a definite answer?"

Eric was at his wit's end. He turned to Seifert and said, "Seifert, you're the CEO now. You make the decision. I can promise to increase the packaging rate of long instructions from 40% to 55%! I suggest we cut our losses now and stop cooperating with Chen Yiran."

Bergman scoffed, "What's the use? The graphics pipeline logic inherited from ATI is outdated! I believe Chen Yiran knows this too. If it weren't for saving our previous shipments, the wisest thing to do would have been to directly target the GCN architecture at the end of the year and simply abandon our previous graphics cards."

Eric roared, "Our shader code is simpler and more organized. It's the most economical way for the compiler to bundle a few independent instructions together and dump them on the hardware, saving the scheduler's transistor and resulting in higher peak floating-point performance."

We can fit more stream processors into the same chip area, so we're just as good as them in pure graphics rendering!

Bergman was also furious, slamming his fist on the table and saying, "All that talk is ultimately just for drawing triangles! We're talking about the future! We're talking about general-purpose computing!"

Otherwise, why would we bother with the GCN architecture? We've brought in outside help precisely to fix the old problems! If you're still clinging to the garbage from the ATI era, you'll never be able to beat Nvidia!

Neither of them could convince the other, and they both looked at Seifert, who was drinking coffee on the side.

Seifert, who had been lost in thought for a while, only looked up when the noise stopped and calmly said, "I don't care who's right or wrong. The contract is signed, so we'll stick to it."

The contract clearly states that he has full technical decision-making authority over the OpenCL compiler. He has the right to propose any modifications within ten working days.

Eric retorted, dissatisfied, "I never expected him to be so radical!"

Seeing that no one was paying attention to him, he angrily stood up, slammed the door, and stormed out, shouting as he left, "You'll regret this!"

Suddenly, without thinking, Xianyu's Seifert exclaimed, "I'm the acting CEO!"

Harry waited until Eric was out of sight before saying, "Shall we send the white paper over?"

Bergman nodded and said, "I'll have someone send it over. According to the previous agreement, I'll also send over the GCN instruction set manual and register specification so that he can prepare for the upgrade and make it easier to port the new framework after it's released."

When Chen Yiran received the document, he felt a little underestimated. He was only asked to prepare for an upgrade; the GCN instruction set manual and register specification were already provided, so all he had to do was write something compatible with both frameworks.

Since his rebirth, he has tried his best to enjoy life, travel around, watch his favorite sports events, raise dogs and pick up girls. With a little help from Jasmine, he has reached a level of technical skill that those mediocre people at AMD cannot comprehend.

"Jasmine, add some more!"

【( > ρ < 」) It's very simple, very simple. Quickly open a glucose bottle, let's get to work!】